House Sale Prediction 2014-2015¶
1.Import, cleaning data and visualization¶
Import libraries
import numpy as np
import pprint
import matplotlib.pyplot as plt
%matplotlib inline
import pandas as pd
import seaborn as sns
from IPython.display import display
import plotly
import plotly.plotly as py
import plotly.graph_objs as go
from plotly.offline import iplot
import cufflinks.offline
import cufflinks as cf
cufflinks.offline.run_from_ipython()
sns.set(style="ticks")
cf.set_config_file(world_readable=True, theme='pearl', offline=True)
plt.rc('font', family='Verdana')
Import data and transform in to pandas dataframe
data = pd.read_csv('/home/shashy/Загрузки/housesalesprediction/kc_house_data.csv')
data.head()
Check the size of dataset
data.shape
Check the info about dataset and search a null objects
data.info()
Some visualization of price
data['price'].iplot(kind='hist', xTitle='price',
yTitle='count', title='Price')
Heatmap
f, ax = plt.subplots(figsize=(16, 12))
corr = data.corr()
sns.heatmap(corr, annot=True)
Visualization price of the houses versus footage ofthe home by date
data.iplot(
x='sqft_living',
y='price',
# Указываем категорию
categories='date',
xTitle='square footage of the home',
yTitle='price of the houses',
title='Price of The Houses vs Square Footage Of The Home by Date ')
delete unnecessary columns
Clear_data_X = data.drop(columns=['id', 'date'], axis=1)
Clear_data_X.head()
Create a discrette scatter to show the relationship between the data
sns.pairplot(Clear_data_X)
2.Splitting + Normalizing data and Predicted Models¶
Separate the target variable from the master data and divide the data into test and training samples.
X = Clear_data_X
y = data['price'].values
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state=42)
After the data is divided, it is necessary to normalize them with the help of a scaler
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
X_train[X_train.columns] = scaler.fit_transform(X_train[X_train.columns])
X_test[X_test.columns] = scaler.fit_transform(X_test[X_test.columns])
We start training our models. For training, I chose 3 models: 1.LinearRegression, 2.RidgeRegression and 3.RandomForrestRegression. As presented below, the linear models and the ensemble of solutions to a random forest coped very well with their work and made a fairly accurate forecast based on test data.
from sklearn.linear_model import LinearRegression
lr = LinearRegression(normalize=True).fit(X_train, y_train)
y_head_lr = lr.predict(X_test)
LinearRegression - 0.99 * 100 = 99% prediction
print("R^2 on train Linear Regression: {:.2f}".format(lr.score(X_train, y_train)))
print("R^2 on test Linear Regression: {:.2f}".format(lr.score(X_test, y_test)))
print("real value of y_test[1]: " + str(y_test[1]) + " -> the predict: " + str(lr.predict(X_test.iloc[[1],:])))
print("real value of y_test[2]: " + str(y_test[2]) + " -> the predict: " + str(lr.predict(X_test.iloc[[2],:])))
RidgeRegression (with params: alpha = 10) - 0.99 * 100 = 99% prediction
from sklearn.linear_model import Ridge
ridge = Ridge(alpha = 10).fit(X_train, y_train)
print("real value of y_test[1]: " + str(y_test[1]) + " -> the predict: " + str(ridge.predict(X_test.iloc[[1],:])))
print("real value of y_test[2]: " + str(y_test[2]) + " -> the predict: " + str(ridge.predict(X_test.iloc[[2],:])))
from sklearn.metrics import r2_score
y_head_ridge = ridge.predict(X_test)
r2score1 = r2_score(y_test, y_head_ridge)
print('R^2 score on ridge: {:.2f}'.format(r2score1))
RandomForrestRegressor (with params: n_jobs = -1, n_estimators = 10, verbose = 3) - 0.99 * 100 = 99% predict
from sklearn.ensemble import RandomForestRegressor
rf = RandomForestRegressor(n_jobs=-1, n_estimators=10, verbose=3)
rf.fit(X_train,y_train)
y_head_rf = rf.predict(X_test)
print("real value of y_test[1]: " + str(y_test[1]) + " -> the predict: " + str(rf.predict(X_test.iloc[[1],:])))
print("real value of y_test[2]: " + str(y_test[2]) + " -> the predict: " + str(rf.predict(X_test.iloc[[2],:])))
print("R^2 score of RandomForrestRegression is: {:.2f}".format(r2_score(y_test, y_head_rf)))
And below is the final accuracy chart of all 3 models.
f, ax = plt.subplots(figsize=(16,10))
y = np.array([r2_score(y_test,y_head_lr),r2_score(y_test,y_head_ridge), r2_score(y_test, y_head_rf)])
x = ["LinearRegression","RidgeReg.", "RandomForrestReg"]
plt.bar(x,y)
plt.title("Accuracy of the models")
Also, I saved the first 100 parameters of each model in the lists. From these lists, you can safely withdraw any value from 1-100
settings = range(1, 101)
lr_predict = []
ridge_predict = []
rf_predict = []
for n_settings in settings:
lr_predict.append("real value of LinearRegression on y_test: " + str(y_test[n_settings]) + " -> the predict: " + str(lr.predict(X_test.iloc[[n_settings],:])))
ridge_predict.append("real value of RidgeRegression on y_test: " + str(y_test[n_settings]) + " -> the predict: " + str(ridge.predict(X_test.iloc[[n_settings],:])))
rf_predict.append("real value of RandomForrestREgression on y_test: " + str(y_test[n_settings]) + " -> the predict: " + str(rf.predict(X_test.iloc[[n_settings],:])))
print(lr_predict[1])
3.Conclusion¶
Thus, we obtained fairly clear machine learning models with an accuracy of 99%. They made a visualization, cleared and divided the data into training and test samples and then normalized the data.